Well, ML on 100 billion searches is probably going to more closely approximate user intent than a living brain in a tank, because that brain doesn't know what the hell "lkw attachment" means.

Looks like german-english bilingual logistics professionals are looking for truck parts vendors, while teenage Americans are looking for hidden locations of laser focusing and enhancement devices within the video game Wolfenstein The New Order.

Do you envision a second route to that answer?

Well, the interface offered by blekko's Izik tablet search engine was that it would show 2 categories in the answer, one related to automotive and one related to games.

Google has done this for a while, but at the level of recognized entities, not at the level of individual queries. Is that how the engine you're referring to worked too? Eg, would the "lkw attachment" example above have partitioned results?

In Izik's tablet interface, each category was a separate row of results. So in this example, there would be 2 rows, one for automotive parts and one for games, and if you scroll horizontally in a row you get more results in that category.

I think that's what you meant by partitioned results.

Google computes this internally but I've never seen them use it for anything other than having diversity in their top 10 results.

Yea it's explicitly separated, just not shown at the same time. For example, if you search for "kings", there will be a couple of bubbles at the top of the page with different entities: "Kings" (2017 film), "Sacramento Kings" (baseball team), etc. Clicking on one of those will show you a list of results that only pertains to that entity. This feature has been around for years, and is part of the series of "things, not strings" features they've been working on.

As I said, Google is pretty conservative about this and other entity-based features, so they definitely wouldn't do it got something like "lkw attachment". My question was whether Izik triggered this feature in such cases or not.

That Google feature is like "related queries", that kind of feature has been around for more than a decade. If you click on the "Kings (2017 Film)" link it runs a search for [Kings 2017]... which just adds 2017 as a keyword to a conventional search. No semantic search is involved.

Izik would show you film-related website results for the film category.